Add timeseries documentation.

content/docs/concepts/timeseries.md

@@ -5,4 +5,28 @@ sort_rank: 1
 
 # Timeseries
 
-TODO: explain how timeseries are identified and stored.
+Prometheus fundamentally stores all data as [timeseries](http://en.wikipedia.org/wiki/Time_series).
+
+## Identification
+Timeseries are uniquely identified by a _metric name_ and a set of _key-value
+pairs_, also known as labels. Changing any label value, including adding or
+removing a label, will result in a new timeseries.
+
+Metric names must match the regular expression `[a-zA-Z_:][a-zA-Z0-9_:]`.
+
+## Samples
+Samples form the actual data that timeseries consist of. A sample consists of:
+
+   * a value with float64-precision
+   * a timestamp with millisecond-precision
+
+## Notation
+Given a metric name and a set of labels, we frequently identify timeseries using this notation:
+
+    <metric name>{<label name>=<label value>, ...}
+
+For example, a timeseries with the metric name `api_http_requests_total` and
+the labels `method="POST"` and `handler="/messages"` could be written like
+this:
+
+    api_http_requests_total{method="POST", handler="/messages"}